Animal Learning & Behavior 1990. 18 (1). 83-89
Categorical color coding in a one-phase avoidance procedure DOMINIC J. ZERBOLIO, JR., and CHRISTIN M. GOLDEN University of Missouri at St. Louis, St. Louis, Missouri Four colors (red, yellow, green, and blue) were arranged in all possible two-color sets to determine if goldfish can discriminate between color sets associated with shock and color sets associated with safety/shock omission in a one-phase (linear presentation) discrimination-learning procedure. The results showed that goldfish learned to discriminate between two-color sets if set colors were natural categorical color-code mates (red = yellow and green = blue). When natural code mates were not in the same set, and therefore were paired with different shock consequents, no discrimination learning occurred, suggesting that goldfish, unlike pigeons, are not able to code colors arbitrarily. The method also allowed a measure of preference between colors within sets associated with safety/shock omission. Original-learning preference measures between colors in sets so associated showed that goldfish chose red over any other color, yellow over blue or green, and green over blue, despite the fact that both colors in any set were procedurally identical, implying that goldfish do discriminate between colors in the absence of explicit discrimination training. The goldfish that failed to discriminate between redlblue and green/yellow sets in original learning were transferred to red/yellow and blue/green color sets. In transfer, the color paired with safety/shock omission in original learning was preferred over the color paired with shock in original learning, which resulted in a reversal of original-learning color preferences for half the goldfish. The transfer color-preference results imply that the goldfish had associated specific colors with specific shock consequents, but the associations were not robust enough to support discrimination learning in the face of categorical color coding. Wright and Cumming (1971) to suggest that pigeons have naturally occurring color categories in which red == yellow and blue == green. Additional support for the notion of natural color-coding categories was found by Zentall, Edwards, Moore, and Hogan (1981). In their study, pigeons that were initially trained on two conditionaldiscrimination tasks, one involving a red/green discrimination and the second a blue/yellow discrimination, continued to perform well when red and yellow were switched or when blue and green were switched (code-mate exchange). However, performance dropped significantly when noncode mates were exchanged (i.e., red and blue or green and yellow). Goldfish in a shock-avoidance procedure were also found to display conditional-discrimination transfer consistent with the CCC phenomenon. When originally trained on a matching conditional-discrirnination task with red/green colors and then transferred to a yellow/blue matching task, transfer was high, but if the transfer task was changed to a yellow/blue oddity task, a reversallearning effect was observed (Zerbolio & Royalty, 1983). Additional support for the CCC phenomenon in goldfish was found using a simple two-color discriminationtransfer task. Goldfish originally trained on a red/green discrimination task were tested in transfer with blue/yellow (Zerbolio, 1985). When red and yellow (and blue and green) were paired the same in the original-learning and transfer phases (both paired with shock, or both paired with safety/shock omission), positive transfer oc-
Because pigeons have color vision, they have been used in behavioral color-discrimination learning studies for the past several decades (Guttman, 1959; Guttman & Kalish, 1956; Honig, 1966; Zentall & Edwards, 1984). Goldfish also have color vision, with retinal spectral sensitivities similar to those of humans (Beauchamp, Rowe, & O'Reilly, 1979; Neumeyer, 1984, 1986; Shefner & Levine, 1976; Yager, 1968). Thus, not surprisingly, goldfish also learn to discriminate behaviorally between any pair of red, green, blue, or yellow colors, even when these colors are produced by simple Christmas-tree lights(Zerbolio, 1981). One of the more counterintuitive behavioral phenomena observed in color-discrimination work is categorical color coding (Ccq. Early observations of the CCC phenomenon occurred with pigeons (Cumming & Berryman, 1961, 1965) in a transfer procedure following initial training on a matching-to-sarnple conditionaldiscrimination. Basically, CCC seems to involve pigeons "coding" yellow == red and green = blue. For example, following the acquisition of a matching response using red/blue colors, if the red sample color is exchanged for yellow in the transfer test, discrimination performance is maintained at high levels. Such findings led Cumming and Berryman (1965) to conclude that pigeons treated the changed sample color as if it were red. These and similar results prompted The authors' mailing address is Department of Psychology, University of Missouri at 51. Louis, 8001 Natural Bridge Road, 51. Louis,
MO 63121.
83
Copyright 1990 Psychonomic Society, Inc.
84
ZERBOLIO AND GOLDEN
curred. If, however, the shock pairing of code-mate colors was changed between original learning (OL) and transfer (T) (red paired with shock on OL and yellow with safety/shock omission in T, or vice versa), an attenuated reversal-learning result was observed. It is clear that most of the pigeon and all of the goldfish results are consistent with Wright and Cumming's (1971) suggestion of (relatively) fixed, naturally occurring color categories, and support their contention of the cross-species generality of the CCC phenomenon. Lack of support for fixed categories comes from Zentall and Edwards's (1984) fmding that pigeons coded red as green and blue as yellow in a transfer test following the initial acquisition of a matching conditional discrimination, indicating some flexibility of which colors may be coded as equivalents. Interestingly enough, Zentall and Edwards (1984) did not find comparable flexibility when initial training and transfer involved a conditional-discrimination oddity task. With pigeons, all demonstrations of CCC have been accomplished with two-phase conditional discriminationtransfer procedures. With goldfish, either conditional discrimination-transfer procedures or two-phase, simple discrimination-transfer procedures have been used. Since all of these procedures require the acquisition of a color discrimination as a necessary precondition before testing for CCC effects, none of them can address the effect of CCC on the original acquisition of a color discrimination. In order to pit the effect of CCC directly against the initial acquisition of a color discrimination, it would be necessary to devise a methodology that allows both CCC and initial learning to occur concurrently. Since discrimination learning occurs in the initial phase of training, by definition, CCC must occur then. This, in principle, necessitates a single-phase, rather than a two-phase, initial-Iearning/transfer-test procedure. Conventional color-discrimination procedures typically present two color stimuli: one color paired with one response-consequent, and the other color paired with the remaining response-consequent. Suppose the single color of the conventional procedure were replaced with a pair, or set, of colors in which either member of the set could occur randomly. Redefining what constitutes a ••stimulus' , as a set of colors allows two different colors to be paired with one response-consequent, and two additional colors to be paired with the remaining response-eonsequent. This would require the presentation of four different colors. With four colors, arranged in sets of two colors for each response-consequent, the task becomes learning to discriminate between sets of colors rather than between colors per se. In addition, only one color of a set should be presented at a time to avoid compound-stimulus interpretations. These methodological requirements can be met for goldfish with a modification of the shuttlebox linear presentation procedure (LPP) (Zerbolio, 1981) developed for this animal. In the original LPP, trial onset was signaled by the offset of a white intertrial interval (IT!) illumination. Each
trial lasted 12.2 sec, with 200-msec shock pulses scheduled to occur at 10 and 12 sec after trial onset. The first shuttle response of the trial, regardless of when it occurred, randomly produced one of two colors. One color (S+) was paired with safety or shock omission, and the second (S-) was paired with shock. The prevailing color alternated with each subsequent response. If the prevailing color at 10 and/or 12 sec was S+, the scheduled shock was omitted. If the prevailing color at 10 and/or 12 sec was S-, shock was delivered as scheduled. If no response occurred within 10 sec of the start of a trial, the shuttlebox remained dark (no color), and shock was delivered as scheduled. The only way for the goldfish to avoid both shocks was to respond before 10 sec, and then, via the onset and alternation contingencies, to control the prevailing color so that the S+ prevailed during both the 10- and the 12-sec shock periods. Any deviation from this requirement resulted in shock being delivered at the first, the second, or both of the scheduled shock times. If, instead of two colors, four colors are used, where color onset or change is controlled by the animal's response, and two of the colors are paired with one response-eonsequent and the other two paired with a second response-consequent, the LPP provides all that is needed to test the effects of CCC against the original acquisition of a discrimination in a single phase. If two code-mated colors formed the first set and were paired with one response-eonsequent (e.g., red and yellow paired with shock), and the second set was composed of the remaining two code-mated colors and as paired with a second response-eonsequent (e.g., blue and green paired with safety/shock omission), one would expect animals to learn to choose between the red/yellow and blue/green color sets fairly easily, and to choose the colors associated with the preferred response-consequent (safety/shock omission). If, in contrast, sets were not formed of codemated colors and, in fact, code mates existed in different sets with different response-consequents (i.e., red/green with one consequent and yellowlblue with the other), one would expect CCC to interfere with, and even prevent, discrimination learning, especially if the natural color categories are fixed. If, however, the natural color categories are flexible, one would expect goldfish to learn to choose the colors in sets associated with the preferred response-consequent independently of natural color categories, although discrimination learning might be a bit slower. In addition to determining if goldfish can learn to choose discriminatively between sets of colors paired with shock and other sets paired with shock omission, the four-color procedure offers the opportunity to see if goldfish choose one color of a set over the other. Choosing one color in a set over the other would imply that goldfish prefer one color over another, even though both members of a set have the same response-consequent. However, a preference measure cannot involve forced or shock-elicitedcolor selection. Measures of color preference between members of a set can occur only on trials
CATEGORICAL COLOR CODING in which no shock occurs, and thus, by definition, only on avoidance trials. In previous work, goldfish have shown idiosyncratic color preferences in a Y-maze avoidance procedure (Zerbolio, 1980) in which both colors led to shock omission. In the LPP, since goldfish typically choosea color and stay with that color choice when it is paired with safety/shock omission (Zerbolio, 1981), preference measures for shock-omission-paired colors in a set wouldbe relatively easy to obtainand interpret. One would notexpect discriminative preferences to develop between the comparably pairedcolors in a set because, as Prokasy and Hall (1963) have suggested, failure to discriminate between even potentially highly discriminable stimuli can occur in the absence of explicit discrimination training. If preferences between colorspaired withshock omission are found, it would imply that goldfish can and do discriminate between colorsin a set, despite the factthatboth colors have the same response-consequent. The goalof the present study was to determine if goldfish canlearnto discriminate between colorsetsassociated withsafety/shock omission and color setsassociated with shock. Since colorsetscouldbe eithercode-mated colors, or noncode mates, the paradigm automatically allows a test of the effect of CCC on the initial acquisition of a color discrimination in a single-phase procedure. In addition,any preferences between colors in sets pairedwith safety/shock omission can be established. METHOD Subjects Ninety-six 5-6-em-long goldfish, obtained from Ozark Fisheries and housed in individual 7.5 x 11.5 x 12.5 em aquariums were used. Water in the aquariums was constantly filtered and aerated. Approximately 25% of the water was siphoned off and replaced daily, which served to remove accumulated solid debris. All aquariums were completely decanted, scrubbed, and refilled weekly. Temperature (21.1 0 C) and pH (7 ± .1) were held constant, and the fish were fed daily. Apparatus Four identical 29.2 x 11.4 x 11.4 ern shuttleboxes were used. A center hurdle, 6.35 cm high with a 9-cm flattop and 45 0 sloping ramps, separated each shuttlebox into deeper wells in each end. Water clearance at 2.2 em over the hurdle top was maintained at all times. Shuttling activity was monitored by two photocells, 9 cm apart, aimed across the flattop center of the hurdle. Photocell light sources were 2.5-V de prefocused penlight bulbs, operated at 1.5-V ac to extend bulb life. Inserts, atttached to the inside end walls of each shuttlebox, provided constant aeration. Colored light sources were created by dipping 12-V ac T-3 1/4 miniature screw lamps into a glass-staining medium ("Great Glass"; Plaid Enterprises, Inc., Norcross, GA). Red (No. 15025), Kelly Green (No. 15009), Royal Blue (No. 15(03), and Yellow (No. 16007) stains were used, producing bulbs with spectral emission peaks of 598 ± 7, 527 ± 7, 484 ± 7, and 546 ± 7 p. as measured with a Spectron Instrument CE395 Fast Spectral Scanner, and referred to in the present study as red, green, blue, and yellow, respectively. Although the spectral wavelengths of the present color sources differed slightly from those used in earlier studies, partial replications of earlier work with the present color sources indicated that goldfish saw
85
them as functional equivalents of the previous red, green, blue, and yellow sources. Two bulbs of each color (red, green, blue, and yellow) were mounted 180 0 apart around the inside diameter (5.2 ern) of a 6.2-cm-Iong tube fixture made of 2-in. white PVC pipe. In addition, two clear bulbs were mounted inside each fixture but were not used in the present study. The tubes were lined with black paper to reduce internal reflection. Identical tube fixtures, centered and mounted about 2 cm from the clear end walls of each shuttlebox, cast colored light into both ends of the shuttlebox. Colored light was spread by diffusion plates placed between the tube fixture and the outside end wall and by the sandblasted aeration inserts inside the ends of each shuttlebox. Brightness of colored lights under water inside the shuttleboxes was measured with a Minolta Spotrneter M. A mirror, resting on the 45 0 sloping ramp of the hurdle, was used to reflect end-color illumination upward. The meter was focused on the mirror from about 5 cm above water level. Measured brightnesses for red, green, blue, and yellow were 8.3±.6, 7.7±.6, 7.7±.8, and 6.2±.4 mL, respectively. An opaque, black, plastic vertical baffle, centered under the hurdle and across the shuttlebox, visually isolated the ends so that a fish in the deep well at one end of the shuttlebox could not see the color of the other end. The underside of each shuttlebox was painted flat black except for a 5-cm clear portion directly under and across the hurdle. ITi illumination was provided by a white, 7-W, IIO-V ac Christmas-tree light centrally placed under the shuttlebox to illuminate the hurdle from underneath. ITi illumination brightness was 6.6±.4 mL. The order of color presented during trials was determined by an EPROM memory device, which functioned as a 96-position stepper, or ring counter, in which all possible permutations (24) of the four colors (red, green, blue, and yellow) were sequentially stored in random order, with the constraint that the same color could not occur in two consecutive positions. Shock was delivered at 7-V ac (.69 V/cm) in 200-msec pulses, via 28 x 10.2 ern, 22-ga stainless steel plates lining the interior side walls of each box. Approximately 25% of the water was siphoned from each box and replaced daily to remove biological waste and debris. Procedure Each experimental condition used four colors (R = red, G = green, B = blue, and Y = yellow), in which two colors were paired with shock (-) and the remaining two colors were paired with shock omission (+). There are three possible ways to arrange the four colors into sets of two colors each: RG/BY, RB/GY, and RY/BG. In addition, a two-color set could be paired with either shock or shock omission, which automatically paired the remaining twocolor set with the remaining shock condition. The three twocolor sets and the two shock possibilities constituted six possible color set x shock combinations: R+G+/B-Y-, R-G-/B+Y+, R+B+/G-Y-, R-B-/G+Y+, R+Y+/B-G-, and R-Y-/B+G+. Note that only the last two of these six combinations have natural color-code mates paired with the same shock consequent (R+Y+/B-G- and R-Y-/B+G+). Sixteen goldfish were run in each of these six possible color set x shock combinations. All goldfish were run for 40 trials a day with a variable 6O-sec ITI. ITIs varied randomly from 30 to 90 sec. The training procedure was a modified form of the LPP described by Zerbolio (1981). Each trial was 12.2 sec long. Trial onset was indicated by the offset of the ITi illumination, rendering the shuttlebox dark. The goldfish's first response on a trial turned on one of the four colors, and each subsequent shuttle during a trial caused a change in the prevailing color, the color being determined by the sequential position of the EPROM device. On average, the same color occurred every 4.0± 1.48 shuttles. Shock, at 7-Vac (.69 V/cm), in 200-msec pulses, was programmed to occur at to and 12 sec after trial onset. To avoid shock,
86
ZERBOLIO AND GOLDEN
the goldfishhad to respond and choose, via its shuttle response(s), oneof thecolors of the set pairedwithshockomission so that,during the programmed shock-delivery periods, one of those colors prevailedin the shuttlebox. If, duringthe programmed shockperiods, any other color (or dark) prevailed, shockwas deliveredas scheduled. Since the prevailing color was tested independently at each scheduled shock period, thegoldfish couldreceive any possible combination of shocks (both, firstonly, second only, or neither) depending on the response-selected color prevailing at scheduled shock times. Each trial was terminated after 12.2 sec, with all color signals turned off and the white IT! lights turned back on. Three responsemeasureswere recordedfor each goldfishat the end of its daily 4O-trial session. Since shock-elicited performance cannot be considered the logical equivalent of performance not forcedby shock, all three indexes excluded forcedor shock-elicited responses. The first measure wasthe number of trialswithresponse (TwR) before 10 sec, whether or not shockwassubsequently delivered. Sinceresponding before 10 sec producedanyone of the four different colors, simply responding did not necessarily produce a shock-omission-paired color, or avoidance. The second measure was the numberof avoidance trials, definedas the numberof TwR trials during whichneitherof the two scheduled shockswas delivered. Bothshockswere omittedonly whenthe goldfishresponded before the first shock (TwR), and then, when necessary, altered the prevailing color with additional responses so that a shockomission-paired color prevailedduring each of the two scheduled shockperiods. The third measure indicated the colorthat prevailed at the end of each avoidance trial. Sincetwo colorswere associated with shock omission, choice of one color over the other implied a preference for the selectedcolor. Earlier two-eolor discrimination work indicated that oncethe preferred(shock-omission-paired) color was chosen,goldfish stayedwiththat color for the remainder of the trial (Zerbolio, 1981). In the present study, on avoidance trials only, the color prevailing at 12.2 sec was measured under theassumption thatit had beenpresentduringtheentire 1O-l2.2-sec period during whichshocks were scheduled to occur, and was defined as the preferredcolor. (Although notquantified, daily observations confirmed this assumption.) In addition, a discrimination index (DI) was calculated for each fish daily. The DI was [(S+)-(S-)]/4O, where S+ is the number of color choices paired with shock omission (avoidances), S- is the number of color choicespaired with shock, S+ and S- equal TwR, and 40 is the number of trials per day. The DI had limits of ± 1.00 (0.00 indicated no discrimination between colors); a positive DI indicated choice of shock-omission-paired colors, and a negative DI indicated choiceof colors paired withshock. Furthermore, for a given proportionof S+ and S- choices, the DI measuredistinguished between an animal's responding on just a few trials or on many trials. Otherresponse measures, suchas m response rate, have not been found to distinguish between experimental groups and were not recorded in the present study. Sincepreferences between colorspairedwithshockomission becameobvious uponcompletion of thefirstphase procedure, webegan to wonder if color preferencewithina set couldbe affected by the reinforcement historiesof the color members of the set. An ad hoc transfer procedure was conducted with the R+B+/G-Y- and R-B-/G+Y+ original-learning groups for which, in transfer, colorswere recombined into natural CCC pairs. The recombination of colors for these groups necessarily producedtransfer-test color sets in which the colors associated with the shock omission had different reinforcement histories. With original-learning/transfer histories indicated by two signs (++, +-, -+, --), groups tested in transfer were R++Y-+/B+-G--, R-+Y++/B--G+-, G-+B++/R+-Y--, and G++B-+/R--Y+- (see Figure 4). With the recombined colors, these groupswere run for 3 days with the same procedure as in original learning to determine if color preferences between the transfer colors paired with shock omis-
sion were the same or different than those observed in original learning.
RESULTS Two-way repeated-measures ANOVAs were computed for the TwR, 01, and avoidance measures with the six, two-color sets x shock-pairing conditions and five, 3-day blocks as factors. Whennecessary, interactions were partitioned to facilitate interpretation. For the TwR measure (see Figure 1), no differences werefound between the sixcolor-set conditions [F(5,9O) = 1.60], but all groups showed reliable increases in TwR over 3-dayblocks of training [F(4,360) = 426.63, p < .01). No interaction was observed. For the 01 measure (see Figure 2), reliable differences between color-set groups [F(5,90) = 27.65, p < .01] and 3-day blocks of training [F(4,360) = 25.32,p < .01], and their interaction [F(20,360) = 6.15, p < .01] were observed. Partitions showed that, for the noncode-mate color sets (RG/BY and RB/GY), no differences due to whichcolor set wasshock-associated wereobserved [F(1,9O) < 1 for both], whereas, for the code-mate set (RY/BG), the R-Y-/B+G+ shock-associated group performed better thantheR+Y+ IB-G- shock-associated group [F(1 ,90) = 11.33, p < .01]. All three color-set groups showed changes over3-day blocks of training [forRGIBY, RYIBG, and RB/GY, Fs(4,360) = 3.97,43.69, and 3.93, respectively, all ps < .01]. Figure 2 shows how the 01 changed with training for all groups. Only the code-mate RY/GB color-set group learned to choose colors associated with safety with days of training and exceeded chance levels in color choiceby the end of training (see Figure 2). The other two color-set groups, with noncode-mate colors (RG/BY and RB/GY), did not learn to choosecolors associated withsafety withtraining and, as Figure 2 shows,
90 80
0::: 70 ~
I- 60 I- 50 Z
W 40
~ ~
30
20 10
t.--I:> YB/RG 0------0
RYI BG
----. RB/GY
o.....:":-_"":"":~-::-'-=-~~---:~~ 1-3 4-6 7-9 10-12 13-15
3-DAY TRAINING BLOCKS Figure 1. Percentage of trials with response (TwR) for color-set groups, shown in 3-day blocks.
CATEGORICAL COLOR CODING
Figure 4 shows the results with the necessary analyses incorporated in the figure. Note that in each case, the color preferred in transfer is thecolorassociated with safety/shock omission in originallearning. Therefore, the original slight, but not reliable, preference for redoveryellow wasaugmented when yellow was paired with shock in original learning, and was reversed when red was paired withshockin original
k----6 Y·B-; R-G-
tr---t:. R-G'; Y-B-
A
_ _ R-Y-; B-G-
0
0 - - 0 B'G'; R-Y-
.3
_ _ R'B'; G-Y-
X
lIJ
0 Z
0---0 GtY-; R-B-
.2
Z
CJ)
~z
070
0
Z
R ~Y
G>B
R>B
~
:E 0
o
ii: o (I)
0::
0
a. 60
o
0::
-.2~
~
4-6
7-9
10-12
13-15
3-DAY TRAINING BLOCKS Figure 2. Discrimination index (»1) measures for coIor-set groups, in ~y blocks. A "+" indicatescolors lIliIlOCiated with safety/shock omission and a "-" indkates colors IL'i!iOCiated with sbock. 00Iy color sets composedof natural color-code mates (see inset box) show evidenceof acqulriDg dB:rlmination performance exteDding consisteDdy outside the .01 cbance-dlscrimination envelope.
were performing at or near chance levels at the end of training. Although the avoidance datawereanalyzed, they added no information not shownby the DI analysis, and are omitted from this report. In addition to theabove analyses, color-preference analyses for the avoidance trials over the 3 days of the last block were performed (see Figure 3). Goldfish could avoidshockby choosing either of the colors paired with safety. If there were no preferences between the safetyrelated colors, one would expect eachto be chosen equally often. Since all possible color set x shock-consequent conditions were represented in the presentdesign, color preferences between eachpossible colorset, in which both members of the set wereassociated withsafety, werecalculatedover the last 3-day blockof training. The following preferences were observed: R=Y [t(I5) = 2.05, p > .05], G>B [t(I5) = 3.84,p < .05], R>B [t(I5) = 3.71, p < .05], Y>B [t(I5) = 4.81, P < .01], R>G [t(I5) = 6.63, p < .01], and Y>G [t(I5) = 9.32, p < .01]. These results show an overall preference order of R = Y > G > B, and with larger groups, it is quite likely that red would be slightly preferred over yellow, but the statistical power of the presentexperiment is not sufficient to confirm this notion. The RB/YG groupswerealsoexamined for 3 days following original learning in a transfer procedure withcolor setscomposed of codemates (i.e., RY/BG). The purpose of this analysis was to compare colorpreference between transfer colors associated with safety/shock omission.
in
,.. ...
a. 50
0
1-3
87
0
•
0
in
CD
-
'""
<5
'"
~40 W
,.;
,..
,..
-
po.
Y>G
R>G
-
0
- -
N
w
-
..;,
Y>B
0
'"
-
'"
0
III
~
oil
N
~
'" ~
0::
~30 W
0::
a.
o
~ 0
z
..J C..J
)
~~
!oJ !oJ !oJ:>
o
'" ..J "CD
'" ..J '" CD
!oJ :>
~ 0
~
o
z
Z ..J\oJ ..J'"
'" '" '" "'"
..J '" ..J :::> '" ..J ~ CD
C\oJ
8+)
"'''' ><:J
(R+G+) (Y+G+) +' (G+ B+) (R+ B+) (Y+ ( R+Y G-B-J R-Y- G-Y- G-R- B-Y- B-R-
COLORS PAIRED WITH SHOCK (-) and SHOCK-OMISSION (+) Figure 3. Comparisons of color preferences for all possible pairs of colors lIliIlOCiated with safety/shock omission (+). Preference proportions, with their lIliIlOCiated t values (d! = IS for each) and their respective p values are incorporated in the figure.
(f)
I
OL
I
I
I
I
I
'"> '" OJ -, °100 Levels I • >• rOJ Z
<,
a:: ~
ft
a. Z 0::
I
70 60
.--
I
f-~
0
W I..L 20 W
a:>-
Q..
00
a:::
c
j
IIJIIJ
10 ..J ..I
0
" l-
~ ~ '" + + >> a::
0
r
rei
I 7I
I C++
~
..J ..J
~~
I
I
Levels I I
I
I
'"I >I > a::
'+ '+ en en + +
a::II '"II I o-' I-
~ ~
I
I g.--
I
OJ I
I
> I
a:: a:: ~
4>
'" lD + + >II II
6
'"
l-
..;,
N
r-
-.... -
I
OL
r-r-
N
I
I
I
o: en
1'--
40 30
+ + a:: a::
I " 6
80
W 50 U
W
I
90
I
,
OJ '" I
0
I
f-
(J>
o ... ~
...
+~ ..J I
C..J
IIJIIJ
a:>-
z
IIJIIJ IIJ ::l
a:..J
(lim
..1..1
6
I~
~
I
I +I ... z + I IIJ LU ::l I IIJ a: ..J
00 I
(lI
-ie
m
t
Z
+ I
IIJIIJ
1IJ::l
a:..J
(lim
ORIGINAL LEARNING and TRANSFER SHOCK PAIRING Figure 4. Transfer (T) color-preference proportions for groups trained in originaIleaming (OL) with RB/GY colors. OL color preferences are shown for comparison purposes. For each group, the OL and T color sets, and theirshock lIf!NllCiatioas, are shown.Speciftc color preferences in transfer depend on the respective originallearning shock-pairing histories (++ or -+) of the colors Involved. The t values and 18 for each are incorporated into the figure.
88
ZERBOLIO AND GOLDEN
learning. A comparable reversal of the green-over-blue preference occurred when green was paired with shock in original learning, but no augmentation of green over blue was seenwhenblue was pairedwith shockin original learning. DISCUSSION The presentresults show that goldfish can learn to discriminate among four colors, when colors are arranged in two-eolor sets, provided that the colors in a set are natural categorical color-code mates (Wright & Cummings, 1971). If the colors in a set are not natural color-code mates or, more importantly, when natural color-eode mates are associated withdifferent shock consequents, no discrimination learning occurs. These data suggest that, unlike pigeons (Zentall & Edwards, 1984), goldfish have no ability to code colors arbitrarily. Thedecision to measure colorpreferences within safetypaired color sets was made after observing goldfish behavior during pilot work. Being aware that goldfish can disriminate between any two of the four colors used, we were, frankly, curious to discover if theirchoices between colors paired with safety were idiosyncratic (Zerbolio, 1980) or followed some heretofore unknown speciesspecific pattern. In a sense, this one measure makes the explanation of the results moredifficult while, at the same time, shedding some light on thepervasiveness of theCCC phenomenon on original discrimination learning. The preference results seemto ruleout anysort of simple generalization explanation for the categorical colorcodingphenomenon in goldfish as the result of a failure to discriminate between or to confuse colors a la Prokasy and Hall (1963). As Figure 3 shows, despite the factthat colors within sets were prograrnatically identical insofar as safety/shock-omission pairing, statistically reliable preferences between them occurred. Clearly, the least preferential differences were between natural category code-mades; red was only slightly (but not reliably) preferred over yellow, and greenwasjust reliably preferred over blue. All other color-preference comparisons were between colors in different natural categories and were highly statistically reliable. It is quiteprobable that even the small red-over-yellow preference would be statistically reliable with more subjects. However, the fmding that any reliable differences in preference occurbetween colorswithin a setthatare comparably consequent-paired argues against a confusion or failure-to-discriminate explanation. The last fmding is a preference reversal in a transfer situation thatwasfound to depend on theoriginal learningconsequent pairing history of the colors in the transfer set. Notethat red was slightly (butnot reliably) preferred over yellow, and green was reliably preferredover blue in the original-learning data(seeFigure 3). Thesepreferences are replicated in Figure 4. Now note what happenedto thesepreferences whenthe R+ B+ /G- Y- and
R-B-/G+Y+ groups were tested in transfer with R+Y+/G-B- or R-Y-/G+B+ color sets. In transfer, when preferences between twosafety/shockomission-paired colors were examined, one color of the set was paired with safety/shock omission in original learning (+ +) andthe othermember of the set waspaired with shock (-+). In all cases, the color member of the set with the + + reinforcement history waspreferred over the remaining member with a -+ history. Thus, when preferences between red andyellow are assessed in transfer, red was significantly preferred over yellow in the R++Y-+ transfer group, andyellow waspreferred over red in the R-+Y++ group, which represents a reversal of the preferences found in original learning. Clearly, the preference between red and yellow is dependent uponthe original-learning shock-pairing history of the specific colorsin the set. The color in the transfer set that had been paired with shock in original learning was least preferred in the transfer situation. To a lesser extent, the samereversal occurred in the B+ +G - + and B-+G++ groups. In the first case, the green-over-blue original-learning preference was reversed, and in the latter, it wasconsistent withthe original-learning preference results. The finding that color preferences can be reversed indicates that goldfish do acquire some colorspecific response-consequent associations in original learning. By themselves, the transfer reversal-of-preference results canbe usedto argueeitherin support of or against Prokasy and Hall's (1963) contention thatdiscrimination does not occur without specific training. Clearly, colors in the transfer setshadbeensubjected to differential reinforcement in original learning. If goldfish had learnedto associate different colors with different shock consequents in original learning, it is not surprising that their color preferences were altered in the transfer situation. But if thatargument is accepted, howdoes oneexplain thatthese same animals never learned to discriminate reliably betweenthese same colors in sets during original learning when the differential shock pairing occurred? One might be tempted to explain these results as a learning/performance difference, that is, animals showing evidence of learning in a transferprocedure that was notpresentin original learning. Hearst(1987), in testing feature-negative discrimination, showed that pigeons had acquired a discrimination between positive and negative cues with an extinction-test procedure after finding little or no evidence for discrimination learning in the original acquisition phase. Unlike Hearst's results, our preference dataindicate thatgoldfish cananddo discriminate between colors in safety-paired color sets and, by implication, probably between colors in shock-paired sets as well, in original learning. Theoriginal-learning colorpreferences cannot be usedas evidence of an acquired discrimination between colors in a set, but are evidence thatgoldfish can discriminate between colors. However, some colorspecific learning mustoccurin original learning because,
CATEGORICAL COLOR CODING
as shown in the transfer-preference data, colorpreferences can be reversed as a function of color-specific associations acquired in original learning. Why, in original learning, goldfish can discriminate between colors in a set, andeven associate specific response-consequences with specific colors but are unable to learn to discriminate between sets of colorsunless the sets are composed of categorical codemates, remains unanswered. It is a puzzlement. The only conclusion that seems reasonable is that, although discrimination learning and categorical colorcoding may be separate processes, the categoricalcolorcoding process is considerably more robust and, in fact, is sufficiently robust to override the discriminationlearning process. Of course, this conclusion must be limited to goldfish. It is quite possible the same conclusion will not fit the behaviorof pigeons. It would not be surprising to find some behavioral and learning differences between phyla. REFERENCES BEAUCHAMP, R. D., ROWE, J. S., '" O'REILLY, L. A. (1979). Goldfish spectral sensitivity: Identification of three cone mechanisms in heart rate conditioning using colored adapting backgrounds. VisionResearch, 19, 1295-1302. CUMMING, W. W., II< BERRYMAN, R. (1961). Some data on matching behavior in the pigeon. Journal of the Experimental Analysis of Behavior, 4, 281-284. CUMMING, W. W., II< BERRYMAN, R. (1965). The complex discriminated operant: Studies of matching-to-sample and related problems. In D. I. Mostofsky (Ed.), Stimulus generalization (pp. 284-330). Stanford, CA: Stanford University Press. GUTTMAN, N. (1959). Generalization gradients around stimuli associated with differentreinforcement schedules. Journal ofExperimental Psychology, 58, 335-340. GUTTMAN, N., II< KALISH, H. I. (1956). Discriminability and stimulus generalization. Journal of Experimental Psychology, 51, 79-88.
89
HEARST, E. (1987). Extinction reveals stimulus control: Latent learning of feature-negative discriminations in pigeons. Journal of Experimental Psychology: Animal Behavior Processes, 13, 52-64. HONIG, W. K. (1966). Operant behavior: Areas of application and research. New York: Appleton-Century-Crofts. NEUMEYER, C. (1984). On spectral sensitivity in the goldfish. Vision Research, 24, 1223-1231. NEUMEYER, C. (1986). Wavelength discrimination in goldfish. Journal of Comparative Physiology A, 158, 203-213. PROKASY, W. F., '" HALL, J. F. (1963). Primary stimulus generalization. Psychological Review, 70, 310-322. SHEFNER, J., '" LEVINE, M. W. (1976). A psychophysical demonstration of goldfish trichromancy. Vision Research, 16, 671-673. WRIGHT, A. A., II
(Manuscript received July 27, 1988; revision accepted for publication April 12, 1989.)